[GFS2] delay glock demote for a minimum hold time
[linux-2.6/kvm.git] / fs / gfs2 / glops.c
blob7ef6b23bb38a9a15dfee3b4a2aff739211eaad1e
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/gfs2_ondisk.h>
15 #include <linux/lm_interface.h>
17 #include "gfs2.h"
18 #include "incore.h"
19 #include "bmap.h"
20 #include "glock.h"
21 #include "glops.h"
22 #include "inode.h"
23 #include "log.h"
24 #include "meta_io.h"
25 #include "recovery.h"
26 #include "rgrp.h"
27 #include "util.h"
28 #include "trans.h"
30 /**
31 * ail_empty_gl - remove all buffers for a given lock from the AIL
32 * @gl: the glock
34 * None of the buffers should be dirty, locked, or pinned.
37 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
39 struct gfs2_sbd *sdp = gl->gl_sbd;
40 unsigned int blocks;
41 struct list_head *head = &gl->gl_ail_list;
42 struct gfs2_bufdata *bd;
43 struct buffer_head *bh;
44 u64 blkno;
45 int error;
47 blocks = atomic_read(&gl->gl_ail_count);
48 if (!blocks)
49 return;
51 error = gfs2_trans_begin(sdp, 0, blocks);
52 if (gfs2_assert_withdraw(sdp, !error))
53 return;
55 gfs2_log_lock(sdp);
56 while (!list_empty(head)) {
57 bd = list_entry(head->next, struct gfs2_bufdata,
58 bd_ail_gl_list);
59 bh = bd->bd_bh;
60 blkno = bh->b_blocknr;
61 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
63 bd->bd_ail = NULL;
64 list_del(&bd->bd_ail_st_list);
65 list_del(&bd->bd_ail_gl_list);
66 atomic_dec(&gl->gl_ail_count);
67 brelse(bh);
68 gfs2_log_unlock(sdp);
70 gfs2_trans_add_revoke(sdp, blkno);
72 gfs2_log_lock(sdp);
74 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
75 gfs2_log_unlock(sdp);
77 gfs2_trans_end(sdp);
78 gfs2_log_flush(sdp, NULL);
81 /**
82 * gfs2_pte_inval - Sync and invalidate all PTEs associated with a glock
83 * @gl: the glock
87 static void gfs2_pte_inval(struct gfs2_glock *gl)
89 struct gfs2_inode *ip;
90 struct inode *inode;
92 ip = gl->gl_object;
93 inode = &ip->i_inode;
94 if (!ip || !S_ISREG(inode->i_mode))
95 return;
97 if (!test_bit(GIF_PAGED, &ip->i_flags))
98 return;
100 unmap_shared_mapping_range(inode->i_mapping, 0, 0);
102 if (test_bit(GIF_SW_PAGED, &ip->i_flags))
103 set_bit(GLF_DIRTY, &gl->gl_flags);
105 clear_bit(GIF_SW_PAGED, &ip->i_flags);
109 * meta_go_sync - sync out the metadata for this glock
110 * @gl: the glock
112 * Called when demoting or unlocking an EX glock. We must flush
113 * to disk all dirty buffers/pages relating to this glock, and must not
114 * not return to caller to demote/unlock the glock until I/O is complete.
117 static void meta_go_sync(struct gfs2_glock *gl)
119 if (gl->gl_state != LM_ST_EXCLUSIVE)
120 return;
122 if (test_and_clear_bit(GLF_DIRTY, &gl->gl_flags)) {
123 gfs2_log_flush(gl->gl_sbd, gl);
124 gfs2_meta_sync(gl);
125 gfs2_ail_empty_gl(gl);
130 * meta_go_inval - invalidate the metadata for this glock
131 * @gl: the glock
132 * @flags:
136 static void meta_go_inval(struct gfs2_glock *gl, int flags)
138 if (!(flags & DIO_METADATA))
139 return;
141 gfs2_meta_inval(gl);
142 gl->gl_vn++;
146 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
147 * @gl: the glock protecting the inode
151 static void inode_go_sync(struct gfs2_glock *gl)
153 struct gfs2_inode *ip = gl->gl_object;
155 if (ip && !S_ISREG(ip->i_inode.i_mode))
156 ip = NULL;
158 if (test_bit(GLF_DIRTY, &gl->gl_flags)) {
159 if (ip && !gfs2_is_jdata(ip))
160 filemap_fdatawrite(ip->i_inode.i_mapping);
161 gfs2_log_flush(gl->gl_sbd, gl);
162 if (ip && gfs2_is_jdata(ip))
163 filemap_fdatawrite(ip->i_inode.i_mapping);
164 gfs2_meta_sync(gl);
165 if (ip) {
166 struct address_space *mapping = ip->i_inode.i_mapping;
167 int error = filemap_fdatawait(mapping);
168 mapping_set_error(mapping, error);
170 clear_bit(GLF_DIRTY, &gl->gl_flags);
171 gfs2_ail_empty_gl(gl);
176 * inode_go_xmote_th - promote/demote a glock
177 * @gl: the glock
178 * @state: the requested state
179 * @flags:
183 static void inode_go_xmote_th(struct gfs2_glock *gl)
185 if (gl->gl_state != LM_ST_UNLOCKED)
186 gfs2_pte_inval(gl);
187 if (gl->gl_state == LM_ST_EXCLUSIVE)
188 inode_go_sync(gl);
192 * inode_go_xmote_bh - After promoting/demoting a glock
193 * @gl: the glock
197 static void inode_go_xmote_bh(struct gfs2_glock *gl)
199 struct gfs2_holder *gh = gl->gl_req_gh;
200 struct buffer_head *bh;
201 int error;
203 if (gl->gl_state != LM_ST_UNLOCKED &&
204 (!gh || !(gh->gh_flags & GL_SKIP))) {
205 error = gfs2_meta_read(gl, gl->gl_name.ln_number, 0, &bh);
206 if (!error)
207 brelse(bh);
212 * inode_go_drop_th - unlock a glock
213 * @gl: the glock
215 * Invoked from rq_demote().
216 * Another node needs the lock in EXCLUSIVE mode, or lock (unused for too long)
217 * is being purged from our node's glock cache; we're dropping lock.
220 static void inode_go_drop_th(struct gfs2_glock *gl)
222 gfs2_pte_inval(gl);
223 if (gl->gl_state == LM_ST_EXCLUSIVE)
224 inode_go_sync(gl);
228 * inode_go_inval - prepare a inode glock to be released
229 * @gl: the glock
230 * @flags:
234 static void inode_go_inval(struct gfs2_glock *gl, int flags)
236 struct gfs2_inode *ip = gl->gl_object;
237 int meta = (flags & DIO_METADATA);
239 if (meta) {
240 gfs2_meta_inval(gl);
241 if (ip)
242 set_bit(GIF_INVALID, &ip->i_flags);
245 if (ip && S_ISREG(ip->i_inode.i_mode)) {
246 truncate_inode_pages(ip->i_inode.i_mapping, 0);
247 clear_bit(GIF_PAGED, &ip->i_flags);
252 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
253 * @gl: the glock
255 * Returns: 1 if it's ok
258 static int inode_go_demote_ok(struct gfs2_glock *gl)
260 struct gfs2_sbd *sdp = gl->gl_sbd;
261 int demote = 0;
263 if (!gl->gl_object && !gl->gl_aspace->i_mapping->nrpages)
264 demote = 1;
265 else if (!sdp->sd_args.ar_localcaching &&
266 time_after_eq(jiffies, gl->gl_stamp +
267 gfs2_tune_get(sdp, gt_demote_secs) * HZ))
268 demote = 1;
270 return demote;
274 * inode_go_lock - operation done after an inode lock is locked by a process
275 * @gl: the glock
276 * @flags:
278 * Returns: errno
281 static int inode_go_lock(struct gfs2_holder *gh)
283 struct gfs2_glock *gl = gh->gh_gl;
284 struct gfs2_inode *ip = gl->gl_object;
285 int error = 0;
287 if (!ip)
288 return 0;
290 if (test_bit(GIF_INVALID, &ip->i_flags)) {
291 error = gfs2_inode_refresh(ip);
292 if (error)
293 return error;
296 if ((ip->i_di.di_flags & GFS2_DIF_TRUNC_IN_PROG) &&
297 (gl->gl_state == LM_ST_EXCLUSIVE) &&
298 (gh->gh_state == LM_ST_EXCLUSIVE))
299 error = gfs2_truncatei_resume(ip);
301 return error;
305 * inode_go_unlock - operation done before an inode lock is unlocked by a
306 * process
307 * @gl: the glock
308 * @flags:
312 static void inode_go_unlock(struct gfs2_holder *gh)
314 struct gfs2_glock *gl = gh->gh_gl;
315 struct gfs2_inode *ip = gl->gl_object;
317 if (ip)
318 gfs2_meta_cache_flush(ip);
322 * rgrp_go_demote_ok - Check to see if it's ok to unlock a RG's glock
323 * @gl: the glock
325 * Returns: 1 if it's ok
328 static int rgrp_go_demote_ok(struct gfs2_glock *gl)
330 return !gl->gl_aspace->i_mapping->nrpages;
334 * rgrp_go_lock - operation done after an rgrp lock is locked by
335 * a first holder on this node.
336 * @gl: the glock
337 * @flags:
339 * Returns: errno
342 static int rgrp_go_lock(struct gfs2_holder *gh)
344 return gfs2_rgrp_bh_get(gh->gh_gl->gl_object);
348 * rgrp_go_unlock - operation done before an rgrp lock is unlocked by
349 * a last holder on this node.
350 * @gl: the glock
351 * @flags:
355 static void rgrp_go_unlock(struct gfs2_holder *gh)
357 gfs2_rgrp_bh_put(gh->gh_gl->gl_object);
361 * trans_go_xmote_th - promote/demote the transaction glock
362 * @gl: the glock
363 * @state: the requested state
364 * @flags:
368 static void trans_go_xmote_th(struct gfs2_glock *gl)
370 struct gfs2_sbd *sdp = gl->gl_sbd;
372 if (gl->gl_state != LM_ST_UNLOCKED &&
373 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
374 gfs2_meta_syncfs(sdp);
375 gfs2_log_shutdown(sdp);
380 * trans_go_xmote_bh - After promoting/demoting the transaction glock
381 * @gl: the glock
385 static void trans_go_xmote_bh(struct gfs2_glock *gl)
387 struct gfs2_sbd *sdp = gl->gl_sbd;
388 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
389 struct gfs2_glock *j_gl = ip->i_gl;
390 struct gfs2_log_header_host head;
391 int error;
393 if (gl->gl_state != LM_ST_UNLOCKED &&
394 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
395 gfs2_meta_cache_flush(GFS2_I(sdp->sd_jdesc->jd_inode));
396 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
398 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
399 if (error)
400 gfs2_consist(sdp);
401 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
402 gfs2_consist(sdp);
404 /* Initialize some head of the log stuff */
405 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
406 sdp->sd_log_sequence = head.lh_sequence + 1;
407 gfs2_log_pointers_init(sdp, head.lh_blkno);
413 * trans_go_drop_th - unlock the transaction glock
414 * @gl: the glock
416 * We want to sync the device even with localcaching. Remember
417 * that localcaching journal replay only marks buffers dirty.
420 static void trans_go_drop_th(struct gfs2_glock *gl)
422 struct gfs2_sbd *sdp = gl->gl_sbd;
424 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
425 gfs2_meta_syncfs(sdp);
426 gfs2_log_shutdown(sdp);
431 * quota_go_demote_ok - Check to see if it's ok to unlock a quota glock
432 * @gl: the glock
434 * Returns: 1 if it's ok
437 static int quota_go_demote_ok(struct gfs2_glock *gl)
439 return !atomic_read(&gl->gl_lvb_count);
442 const struct gfs2_glock_operations gfs2_meta_glops = {
443 .go_xmote_th = meta_go_sync,
444 .go_drop_th = meta_go_sync,
445 .go_type = LM_TYPE_META,
448 const struct gfs2_glock_operations gfs2_inode_glops = {
449 .go_xmote_th = inode_go_xmote_th,
450 .go_xmote_bh = inode_go_xmote_bh,
451 .go_drop_th = inode_go_drop_th,
452 .go_inval = inode_go_inval,
453 .go_demote_ok = inode_go_demote_ok,
454 .go_lock = inode_go_lock,
455 .go_unlock = inode_go_unlock,
456 .go_type = LM_TYPE_INODE,
457 .go_min_hold_time = HZ / 10,
460 const struct gfs2_glock_operations gfs2_rgrp_glops = {
461 .go_xmote_th = meta_go_sync,
462 .go_drop_th = meta_go_sync,
463 .go_inval = meta_go_inval,
464 .go_demote_ok = rgrp_go_demote_ok,
465 .go_lock = rgrp_go_lock,
466 .go_unlock = rgrp_go_unlock,
467 .go_type = LM_TYPE_RGRP,
468 .go_min_hold_time = HZ / 10,
471 const struct gfs2_glock_operations gfs2_trans_glops = {
472 .go_xmote_th = trans_go_xmote_th,
473 .go_xmote_bh = trans_go_xmote_bh,
474 .go_drop_th = trans_go_drop_th,
475 .go_type = LM_TYPE_NONDISK,
478 const struct gfs2_glock_operations gfs2_iopen_glops = {
479 .go_type = LM_TYPE_IOPEN,
482 const struct gfs2_glock_operations gfs2_flock_glops = {
483 .go_type = LM_TYPE_FLOCK,
486 const struct gfs2_glock_operations gfs2_nondisk_glops = {
487 .go_type = LM_TYPE_NONDISK,
490 const struct gfs2_glock_operations gfs2_quota_glops = {
491 .go_demote_ok = quota_go_demote_ok,
492 .go_type = LM_TYPE_QUOTA,
495 const struct gfs2_glock_operations gfs2_journal_glops = {
496 .go_type = LM_TYPE_JOURNAL,